memcg: synchronized LRU
authorKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Thu, 8 Jan 2009 02:08:01 +0000 (18:08 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 8 Jan 2009 16:31:05 +0000 (08:31 -0800)
A big patch for changing memcg's LRU semantics.

Now,
  - page_cgroup is linked to mem_cgroup's its own LRU (per zone).

  - LRU of page_cgroup is not synchronous with global LRU.

  - page and page_cgroup is one-to-one and statically allocated.

  - To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
    - lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);

  - SwapCache is handled.

And, when we handle LRU list of page_cgroup, we do following.

pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);

But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.

This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as

        spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
        spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU

This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
    1. When pc->mem_cgroup can be modified.
       - at charge.
       - at account_move().
    2. at charge
       the PCG_USED bit is not set before pc->mem_cgroup is fixed.
    3. at account_move()
       the page is isolated and not on LRU.

Pros.
  - easy for maintenance.
  - memcg can make use of laziness of pagevec.
  - we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
  - LRU status of memcg will be synchronized with global LRU's one.
  - # of locks are reduced.
  - account_move() is simplified very much.
Cons.
  - may increase cost of LRU rotation.
    (no impact if memcg is not configured.)

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/splice.c
include/linux/memcontrol.h
include/linux/mm_inline.h
include/linux/page_cgroup.h
mm/memcontrol.c
mm/page_cgroup.c
mm/swap.c
mm/vmscan.c

index 1abab5cee4ba8d552df8300e3561ff7e8a1375c3..a54b3e3f10a716cb63ab56311745716a70086d0f 100644 (file)
@@ -21,6 +21,7 @@
 #include <linux/file.h>
 #include <linux/pagemap.h>
 #include <linux/splice.h>
+#include <linux/memcontrol.h>
 #include <linux/mm_inline.h>
 #include <linux/swap.h>
 #include <linux/writeback.h>
index ca51ac72d6c0f7d77e0b6d5ca64aeaa2b343ec5d..32c07b1852d6f7db0df1cff358f3ad47a20ce667 100644 (file)
@@ -40,7 +40,12 @@ extern void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *ptr);
 
 extern int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
                                        gfp_t gfp_mask);
-extern void mem_cgroup_move_lists(struct page *page, enum lru_list lru);
+extern void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru);
+extern void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru);
+extern void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru);
+extern void mem_cgroup_del_lru(struct page *page);
+extern void mem_cgroup_move_lists(struct page *page,
+                                 enum lru_list from, enum lru_list to);
 extern void mem_cgroup_uncharge_page(struct page *page);
 extern void mem_cgroup_uncharge_cache_page(struct page *page);
 extern int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask);
@@ -131,7 +136,27 @@ static inline int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
        return 0;
 }
 
-static inline void mem_cgroup_move_lists(struct page *page, bool active)
+static inline void mem_cgroup_add_lru_list(struct page *page, int lru)
+{
+}
+
+static inline void mem_cgroup_del_lru_list(struct page *page, int lru)
+{
+       return ;
+}
+
+static inline void mem_cgroup_rotate_lru_list(struct page *page, int lru)
+{
+       return ;
+}
+
+static inline void mem_cgroup_del_lru(struct page *page)
+{
+       return ;
+}
+
+static inline void
+mem_cgroup_move_lists(struct page *page, enum lru_list from, enum lru_list to)
 {
 }
 
index c948350c378e93cb9144e9c17c52f3445e8f4a80..37ef13d0f01ecb7fab1901274348020b51c8e51c 100644 (file)
@@ -28,6 +28,7 @@ add_page_to_lru_list(struct zone *zone, struct page *page, enum lru_list l)
 {
        list_add(&page->lru, &zone->lru[l].list);
        __inc_zone_state(zone, NR_LRU_BASE + l);
+       mem_cgroup_add_lru_list(page, l);
 }
 
 static inline void
@@ -35,6 +36,7 @@ del_page_from_lru_list(struct zone *zone, struct page *page, enum lru_list l)
 {
        list_del(&page->lru);
        __dec_zone_state(zone, NR_LRU_BASE + l);
+       mem_cgroup_del_lru_list(page, l);
 }
 
 static inline void
@@ -54,6 +56,7 @@ del_page_from_lru(struct zone *zone, struct page *page)
                l += page_is_file_cache(page);
        }
        __dec_zone_state(zone, NR_LRU_BASE + l);
+       mem_cgroup_del_lru_list(page, l);
 }
 
 /**
index d754b2dfbf2d6429635a2ebebd0579d5a43fa804..602cc1fdee90c152dc458642256b623ce8cbfa78 100644 (file)
@@ -26,10 +26,6 @@ enum {
        PCG_LOCK,  /* page cgroup is locked */
        PCG_CACHE, /* charged as cache */
        PCG_USED, /* this object is in use. */
-       /* flags for LRU placement */
-       PCG_ACTIVE, /* page is active in this cgroup */
-       PCG_FILE, /* page is file system backed */
-       PCG_UNEVICTABLE, /* page is unevictableable */
 };
 
 #define TESTPCGFLAG(uname, lname)                      \
@@ -50,19 +46,6 @@ TESTPCGFLAG(Cache, CACHE)
 TESTPCGFLAG(Used, USED)
 CLEARPCGFLAG(Used, USED)
 
-/* LRU management flags (from global-lru definition) */
-TESTPCGFLAG(File, FILE)
-SETPCGFLAG(File, FILE)
-CLEARPCGFLAG(File, FILE)
-
-TESTPCGFLAG(Active, ACTIVE)
-SETPCGFLAG(Active, ACTIVE)
-CLEARPCGFLAG(Active, ACTIVE)
-
-TESTPCGFLAG(Unevictable, UNEVICTABLE)
-SETPCGFLAG(Unevictable, UNEVICTABLE)
-CLEARPCGFLAG(Unevictable, UNEVICTABLE)
-
 static inline int page_cgroup_nid(struct page_cgroup *pc)
 {
        return page_to_nid(pc->page);
index 2efcf38f3b73b18ef6040b607ad956c404c0d5b0..8ce4e9e4795960e5258f3c77424fb209094b9d7a 100644 (file)
@@ -36,6 +36,7 @@
 #include <linux/vmalloc.h>
 #include <linux/mm_inline.h>
 #include <linux/page_cgroup.h>
+#include "internal.h"
 
 #include <asm/uaccess.h>
 
@@ -100,7 +101,6 @@ struct mem_cgroup_per_zone {
        /*
         * spin_lock to protect the per cgroup LRU
         */
-       spinlock_t              lru_lock;
        struct list_head        lists[NR_LRU_LISTS];
        unsigned long           count[NR_LRU_LISTS];
 };
@@ -163,14 +163,12 @@ enum charge_type {
 /* only for here (for easy reading.) */
 #define PCGF_CACHE     (1UL << PCG_CACHE)
 #define PCGF_USED      (1UL << PCG_USED)
-#define PCGF_ACTIVE    (1UL << PCG_ACTIVE)
 #define PCGF_LOCK      (1UL << PCG_LOCK)
-#define PCGF_FILE      (1UL << PCG_FILE)
 static const unsigned long
 pcg_default_flags[NR_CHARGE_TYPE] = {
-       PCGF_CACHE | PCGF_FILE | PCGF_USED | PCGF_LOCK, /* File Cache */
-       PCGF_ACTIVE | PCGF_USED | PCGF_LOCK, /* Anon */
-       PCGF_ACTIVE | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
+       PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
+       PCGF_USED | PCGF_LOCK, /* Anon */
+       PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
        0, /* FORCE */
 };
 
@@ -185,9 +183,6 @@ pcg_default_flags[NR_CHARGE_TYPE] = {
 static void mem_cgroup_get(struct mem_cgroup *mem);
 static void mem_cgroup_put(struct mem_cgroup *mem);
 
-/*
- * Always modified under lru lock. Then, not necessary to preempt_disable()
- */
 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
                                         struct page_cgroup *pc,
                                         bool charge)
@@ -195,10 +190,9 @@ static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
        int val = (charge)? 1 : -1;
        struct mem_cgroup_stat *stat = &mem->stat;
        struct mem_cgroup_stat_cpu *cpustat;
+       int cpu = get_cpu();
 
-       VM_BUG_ON(!irqs_disabled());
-
-       cpustat = &stat->cpustat[smp_processor_id()];
+       cpustat = &stat->cpustat[cpu];
        if (PageCgroupCache(pc))
                __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
        else
@@ -210,6 +204,7 @@ static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
        else
                __mem_cgroup_stat_add_safe(cpustat,
                                MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
+       put_cpu();
 }
 
 static struct mem_cgroup_per_zone *
@@ -264,80 +259,95 @@ struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
                                struct mem_cgroup, css);
 }
 
-static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz,
-                       struct page_cgroup *pc)
-{
-       int lru = LRU_BASE;
+/*
+ * Following LRU functions are allowed to be used without PCG_LOCK.
+ * Operations are called by routine of global LRU independently from memcg.
+ * What we have to take care of here is validness of pc->mem_cgroup.
+ *
+ * Changes to pc->mem_cgroup happens when
+ * 1. charge
+ * 2. moving account
+ * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
+ * It is added to LRU before charge.
+ * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
+ * When moving account, the page is not on LRU. It's isolated.
+ */
 
-       if (PageCgroupUnevictable(pc))
-               lru = LRU_UNEVICTABLE;
-       else {
-               if (PageCgroupActive(pc))
-                       lru += LRU_ACTIVE;
-               if (PageCgroupFile(pc))
-                       lru += LRU_FILE;
-       }
+void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
+{
+       struct page_cgroup *pc;
+       struct mem_cgroup *mem;
+       struct mem_cgroup_per_zone *mz;
 
+       if (mem_cgroup_subsys.disabled)
+               return;
+       pc = lookup_page_cgroup(page);
+       /* can happen while we handle swapcache. */
+       if (list_empty(&pc->lru))
+               return;
+       mz = page_cgroup_zoneinfo(pc);
+       mem = pc->mem_cgroup;
        MEM_CGROUP_ZSTAT(mz, lru) -= 1;
-
-       mem_cgroup_charge_statistics(pc->mem_cgroup, pc, false);
-       list_del(&pc->lru);
+       list_del_init(&pc->lru);
+       return;
 }
 
-static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz,
-                               struct page_cgroup *pc, bool hot)
+void mem_cgroup_del_lru(struct page *page)
 {
-       int lru = LRU_BASE;
+       mem_cgroup_del_lru_list(page, page_lru(page));
+}
 
-       if (PageCgroupUnevictable(pc))
-               lru = LRU_UNEVICTABLE;
-       else {
-               if (PageCgroupActive(pc))
-                       lru += LRU_ACTIVE;
-               if (PageCgroupFile(pc))
-                       lru += LRU_FILE;
-       }
+void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
+{
+       struct mem_cgroup_per_zone *mz;
+       struct page_cgroup *pc;
 
-       MEM_CGROUP_ZSTAT(mz, lru) += 1;
-       if (hot)
-               list_add(&pc->lru, &mz->lists[lru]);
-       else
-               list_add_tail(&pc->lru, &mz->lists[lru]);
+       if (mem_cgroup_subsys.disabled)
+               return;
 
-       mem_cgroup_charge_statistics(pc->mem_cgroup, pc, true);
+       pc = lookup_page_cgroup(page);
+       smp_rmb();
+       /* unused page is not rotated. */
+       if (!PageCgroupUsed(pc))
+               return;
+       mz = page_cgroup_zoneinfo(pc);
+       list_move(&pc->lru, &mz->lists[lru]);
 }
 
-static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru)
+void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
 {
-       struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
-       int active    = PageCgroupActive(pc);
-       int file      = PageCgroupFile(pc);
-       int unevictable = PageCgroupUnevictable(pc);
-       enum lru_list from = unevictable ? LRU_UNEVICTABLE :
-                               (LRU_FILE * !!file + !!active);
+       struct page_cgroup *pc;
+       struct mem_cgroup_per_zone *mz;
 
-       if (lru == from)
+       if (mem_cgroup_subsys.disabled)
+               return;
+       pc = lookup_page_cgroup(page);
+       /* barrier to sync with "charge" */
+       smp_rmb();
+       if (!PageCgroupUsed(pc))
                return;
 
-       MEM_CGROUP_ZSTAT(mz, from) -= 1;
-       /*
-        * However this is done under mz->lru_lock, another flags, which
-        * are not related to LRU, will be modified from out-of-lock.
-        * We have to use atomic set/clear flags.
-        */
-       if (is_unevictable_lru(lru)) {
-               ClearPageCgroupActive(pc);
-               SetPageCgroupUnevictable(pc);
-       } else {
-               if (is_active_lru(lru))
-                       SetPageCgroupActive(pc);
-               else
-                       ClearPageCgroupActive(pc);
-               ClearPageCgroupUnevictable(pc);
-       }
-
+       mz = page_cgroup_zoneinfo(pc);
        MEM_CGROUP_ZSTAT(mz, lru) += 1;
-       list_move(&pc->lru, &mz->lists[lru]);
+       list_add(&pc->lru, &mz->lists[lru]);
+}
+/*
+ * To add swapcache into LRU. Be careful to all this function.
+ * zone->lru_lock shouldn't be held and irq must not be disabled.
+ */
+static void mem_cgroup_lru_fixup(struct page *page)
+{
+       if (!isolate_lru_page(page))
+               putback_lru_page(page);
+}
+
+void mem_cgroup_move_lists(struct page *page,
+                          enum lru_list from, enum lru_list to)
+{
+       if (mem_cgroup_subsys.disabled)
+               return;
+       mem_cgroup_del_lru_list(page, from);
+       mem_cgroup_add_lru_list(page, to);
 }
 
 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
@@ -350,37 +360,6 @@ int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
        return ret;
 }
 
-/*
- * This routine assumes that the appropriate zone's lru lock is already held
- */
-void mem_cgroup_move_lists(struct page *page, enum lru_list lru)
-{
-       struct page_cgroup *pc;
-       struct mem_cgroup_per_zone *mz;
-       unsigned long flags;
-
-       if (mem_cgroup_subsys.disabled)
-               return;
-
-       /*
-        * We cannot lock_page_cgroup while holding zone's lru_lock,
-        * because other holders of lock_page_cgroup can be interrupted
-        * with an attempt to rotate_reclaimable_page.  But we cannot
-        * safely get to page_cgroup without it, so just try_lock it:
-        * mem_cgroup_isolate_pages allows for page left on wrong list.
-        */
-       pc = lookup_page_cgroup(page);
-       if (!trylock_page_cgroup(pc))
-               return;
-       if (pc && PageCgroupUsed(pc)) {
-               mz = page_cgroup_zoneinfo(pc);
-               spin_lock_irqsave(&mz->lru_lock, flags);
-               __mem_cgroup_move_lists(pc, lru);
-               spin_unlock_irqrestore(&mz->lru_lock, flags);
-       }
-       unlock_page_cgroup(pc);
-}
-
 /*
  * Calculate mapped_ratio under memory controller. This will be used in
  * vmscan.c for deteremining we have to reclaim mapped pages.
@@ -460,40 +439,24 @@ unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
        mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
        src = &mz->lists[lru];
 
-       spin_lock(&mz->lru_lock);
        scan = 0;
        list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
                if (scan >= nr_to_scan)
                        break;
+
+               page = pc->page;
                if (unlikely(!PageCgroupUsed(pc)))
                        continue;
-               page = pc->page;
-
                if (unlikely(!PageLRU(page)))
                        continue;
 
-               /*
-                * TODO: play better with lumpy reclaim, grabbing anything.
-                */
-               if (PageUnevictable(page) ||
-                   (PageActive(page) && !active) ||
-                   (!PageActive(page) && active)) {
-                       __mem_cgroup_move_lists(pc, page_lru(page));
-                       continue;
-               }
-
                scan++;
-               list_move(&pc->lru, &pc_list);
-
                if (__isolate_lru_page(page, mode, file) == 0) {
                        list_move(&page->lru, dst);
                        nr_taken++;
                }
        }
 
-       list_splice(&pc_list, src);
-       spin_unlock(&mz->lru_lock);
-
        *scanned = scan;
        return nr_taken;
 }
@@ -608,9 +571,6 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
                                     struct page_cgroup *pc,
                                     enum charge_type ctype)
 {
-       struct mem_cgroup_per_zone *mz;
-       unsigned long flags;
-
        /* try_charge() can return NULL to *memcg, taking care of it. */
        if (!mem)
                return;
@@ -625,17 +585,11 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
                return;
        }
        pc->mem_cgroup = mem;
-       /*
-        * If a page is accounted as a page cache, insert to inactive list.
-        * If anon, insert to active list.
-        */
+       smp_wmb();
        pc->flags = pcg_default_flags[ctype];
 
-       mz = page_cgroup_zoneinfo(pc);
+       mem_cgroup_charge_statistics(mem, pc, true);
 
-       spin_lock_irqsave(&mz->lru_lock, flags);
-       __mem_cgroup_add_list(mz, pc, true);
-       spin_unlock_irqrestore(&mz->lru_lock, flags);
        unlock_page_cgroup(pc);
 }
 
@@ -646,8 +600,7 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
  * @to:        mem_cgroup which the page is moved to. @from != @to.
  *
  * The caller must confirm following.
- * 1. disable irq.
- * 2. lru_lock of old mem_cgroup(@from) should be held.
+ * - page is not on LRU (isolate_page() is useful.)
  *
  * returns 0 at success,
  * returns -EBUSY when lock is busy or "pc" is unstable.
@@ -663,15 +616,14 @@ static int mem_cgroup_move_account(struct page_cgroup *pc,
        int nid, zid;
        int ret = -EBUSY;
 
-       VM_BUG_ON(!irqs_disabled());
        VM_BUG_ON(from == to);
+       VM_BUG_ON(PageLRU(pc->page));
 
        nid = page_cgroup_nid(pc);
        zid = page_cgroup_zid(pc);
        from_mz =  mem_cgroup_zoneinfo(from, nid, zid);
        to_mz =  mem_cgroup_zoneinfo(to, nid, zid);
 
-
        if (!trylock_page_cgroup(pc))
                return ret;
 
@@ -681,18 +633,15 @@ static int mem_cgroup_move_account(struct page_cgroup *pc,
        if (pc->mem_cgroup != from)
                goto out;
 
-       if (spin_trylock(&to_mz->lru_lock)) {
-               __mem_cgroup_remove_list(from_mz, pc);
-               css_put(&from->css);
-               res_counter_uncharge(&from->res, PAGE_SIZE);
-               if (do_swap_account)
-                       res_counter_uncharge(&from->memsw, PAGE_SIZE);
-               pc->mem_cgroup = to;
-               css_get(&to->css);
-               __mem_cgroup_add_list(to_mz, pc, false);
-               ret = 0;
-               spin_unlock(&to_mz->lru_lock);
-       }
+       css_put(&from->css);
+       res_counter_uncharge(&from->res, PAGE_SIZE);
+       mem_cgroup_charge_statistics(from, pc, false);
+       if (do_swap_account)
+               res_counter_uncharge(&from->memsw, PAGE_SIZE);
+       pc->mem_cgroup = to;
+       mem_cgroup_charge_statistics(to, pc, true);
+       css_get(&to->css);
+       ret = 0;
 out:
        unlock_page_cgroup(pc);
        return ret;
@@ -706,39 +655,47 @@ static int mem_cgroup_move_parent(struct page_cgroup *pc,
                                  struct mem_cgroup *child,
                                  gfp_t gfp_mask)
 {
+       struct page *page = pc->page;
        struct cgroup *cg = child->css.cgroup;
        struct cgroup *pcg = cg->parent;
        struct mem_cgroup *parent;
-       struct mem_cgroup_per_zone *mz;
-       unsigned long flags;
        int ret;
 
        /* Is ROOT ? */
        if (!pcg)
                return -EINVAL;
 
+
        parent = mem_cgroup_from_cont(pcg);
 
+
        ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
        if (ret)
                return ret;
 
-       mz = mem_cgroup_zoneinfo(child,
-                       page_cgroup_nid(pc), page_cgroup_zid(pc));
+       if (!get_page_unless_zero(page))
+               return -EBUSY;
+
+       ret = isolate_lru_page(page);
+
+       if (ret)
+               goto cancel;
 
-       spin_lock_irqsave(&mz->lru_lock, flags);
        ret = mem_cgroup_move_account(pc, child, parent);
-       spin_unlock_irqrestore(&mz->lru_lock, flags);
 
-       /* drop extra refcnt */
+       /* drop extra refcnt by try_charge() (move_account increment one) */
        css_put(&parent->css);
-       /* uncharge if move fails */
-       if (ret) {
-               res_counter_uncharge(&parent->res, PAGE_SIZE);
-               if (do_swap_account)
-                       res_counter_uncharge(&parent->memsw, PAGE_SIZE);
+       putback_lru_page(page);
+       if (!ret) {
+               put_page(page);
+               return 0;
        }
-
+       /* uncharge if move fails */
+cancel:
+       res_counter_uncharge(&parent->res, PAGE_SIZE);
+       if (do_swap_account)
+               res_counter_uncharge(&parent->memsw, PAGE_SIZE);
+       put_page(page);
        return ret;
 }
 
@@ -912,6 +869,8 @@ int mem_cgroup_cache_charge_swapin(struct page *page,
        }
        if (!locked)
                unlock_page(page);
+       /* add this page(page_cgroup) to the LRU we want. */
+       mem_cgroup_lru_fixup(page);
 
        return ret;
 }
@@ -944,6 +903,8 @@ void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
                }
 
        }
+       /* add this page(page_cgroup) to the LRU we want. */
+       mem_cgroup_lru_fixup(page);
 }
 
 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
@@ -968,7 +929,6 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
        struct page_cgroup *pc;
        struct mem_cgroup *mem = NULL;
        struct mem_cgroup_per_zone *mz;
-       unsigned long flags;
 
        if (mem_cgroup_subsys.disabled)
                return NULL;
@@ -1010,12 +970,10 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
        if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
                res_counter_uncharge(&mem->memsw, PAGE_SIZE);
 
+       mem_cgroup_charge_statistics(mem, pc, false);
        ClearPageCgroupUsed(pc);
 
        mz = page_cgroup_zoneinfo(pc);
-       spin_lock_irqsave(&mz->lru_lock, flags);
-       __mem_cgroup_remove_list(mz, pc);
-       spin_unlock_irqrestore(&mz->lru_lock, flags);
        unlock_page_cgroup(pc);
 
        css_put(&mem->css);
@@ -1281,21 +1239,22 @@ int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
        return ret;
 }
 
-
 /*
  * This routine traverse page_cgroup in given list and drop them all.
  * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
  */
 static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
-                           struct mem_cgroup_per_zone *mz,
-                           enum lru_list lru)
+                               int node, int zid, enum lru_list lru)
 {
+       struct zone *zone;
+       struct mem_cgroup_per_zone *mz;
        struct page_cgroup *pc, *busy;
-       unsigned long flags;
-       unsigned long loop;
+       unsigned long flags, loop;
        struct list_head *list;
        int ret = 0;
 
+       zone = &NODE_DATA(node)->node_zones[zid];
+       mz = mem_cgroup_zoneinfo(mem, node, zid);
        list = &mz->lists[lru];
 
        loop = MEM_CGROUP_ZSTAT(mz, lru);
@@ -1304,19 +1263,19 @@ static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
        busy = NULL;
        while (loop--) {
                ret = 0;
-               spin_lock_irqsave(&mz->lru_lock, flags);
+               spin_lock_irqsave(&zone->lru_lock, flags);
                if (list_empty(list)) {
-                       spin_unlock_irqrestore(&mz->lru_lock, flags);
+                       spin_unlock_irqrestore(&zone->lru_lock, flags);
                        break;
                }
                pc = list_entry(list->prev, struct page_cgroup, lru);
                if (busy == pc) {
                        list_move(&pc->lru, list);
                        busy = 0;
-                       spin_unlock_irqrestore(&mz->lru_lock, flags);
+                       spin_unlock_irqrestore(&zone->lru_lock, flags);
                        continue;
                }
-               spin_unlock_irqrestore(&mz->lru_lock, flags);
+               spin_unlock_irqrestore(&zone->lru_lock, flags);
 
                ret = mem_cgroup_move_parent(pc, mem, GFP_HIGHUSER_MOVABLE);
                if (ret == -ENOMEM)
@@ -1329,6 +1288,7 @@ static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
                } else
                        busy = NULL;
        }
+
        if (!ret && !list_empty(list))
                return -EBUSY;
        return ret;
@@ -1364,12 +1324,10 @@ move_account:
                ret = 0;
                for_each_node_state(node, N_POSSIBLE) {
                        for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
-                               struct mem_cgroup_per_zone *mz;
                                enum lru_list l;
-                               mz = mem_cgroup_zoneinfo(mem, node, zid);
                                for_each_lru(l) {
                                        ret = mem_cgroup_force_empty_list(mem,
-                                                                 mz, l);
+                                                       node, zid, l);
                                        if (ret)
                                                break;
                                }
@@ -1413,6 +1371,7 @@ try_to_free:
                }
 
        }
+       lru_add_drain();
        /* try move_account...there may be some *locked* pages. */
        if (mem->res.usage)
                goto move_account;
@@ -1657,7 +1616,6 @@ static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
 
        for (zone = 0; zone < MAX_NR_ZONES; zone++) {
                mz = &pn->zoneinfo[zone];
-               spin_lock_init(&mz->lru_lock);
                for_each_lru(l)
                        INIT_LIST_HEAD(&mz->lists[l]);
        }
@@ -1706,8 +1664,15 @@ static struct mem_cgroup *mem_cgroup_alloc(void)
 
 static void mem_cgroup_free(struct mem_cgroup *mem)
 {
+       int node;
+
        if (atomic_read(&mem->refcnt) > 0)
                return;
+
+
+       for_each_node_state(node, N_POSSIBLE)
+               free_mem_cgroup_per_zone_info(mem, node);
+
        if (mem_cgroup_size() < PAGE_SIZE)
                kfree(mem);
        else
@@ -1780,12 +1745,6 @@ static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
                                struct cgroup *cont)
 {
-       int node;
-       struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
-
-       for_each_node_state(node, N_POSSIBLE)
-               free_mem_cgroup_per_zone_info(mem, node);
-
        mem_cgroup_free(mem_cgroup_from_cont(cont));
 }
 
index 685e7c8e1fd631123f2c7093b943dbb38c81c31a..74ae8e01d071492ddedd53909e397c1167a52c6b 100644 (file)
@@ -16,6 +16,7 @@ __init_page_cgroup(struct page_cgroup *pc, unsigned long pfn)
        pc->flags = 0;
        pc->mem_cgroup = NULL;
        pc->page = pfn_to_page(pfn);
+       INIT_LIST_HEAD(&pc->lru);
 }
 static unsigned long total_usage;
 
index ba2c0e8b8b54387f1743ee60f3d97301820f1b28..8a98a9c9070470b78247920a02dd3666017b1c16 100644 (file)
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -168,7 +168,6 @@ void activate_page(struct page *page)
                lru += LRU_ACTIVE;
                add_page_to_lru_list(zone, page, lru);
                __count_vm_event(PGACTIVATE);
-               mem_cgroup_move_lists(page, lru);
 
                zone->recent_rotated[!!file]++;
                zone->recent_scanned[!!file]++;
index f63b20dd7714989f2f2fe389cf81771358b15d1d..45983af1de3d5be02dc46119213426f72847a4da 100644 (file)
@@ -512,7 +512,6 @@ redo:
                lru = LRU_UNEVICTABLE;
                add_page_to_unevictable_list(page);
        }
-       mem_cgroup_move_lists(page, lru);
 
        /*
         * page's status can change while we move it among lru. If an evictable
@@ -547,7 +546,6 @@ void putback_lru_page(struct page *page)
 
        lru = !!TestClearPageActive(page) + page_is_file_cache(page);
        lru_cache_add_lru(page, lru);
-       mem_cgroup_move_lists(page, lru);
        put_page(page);
 }
 #endif /* CONFIG_UNEVICTABLE_LRU */
@@ -813,6 +811,7 @@ int __isolate_lru_page(struct page *page, int mode, int file)
                return ret;
 
        ret = -EBUSY;
+
        if (likely(get_page_unless_zero(page))) {
                /*
                 * Be careful not to clear PageLRU until after we're
@@ -821,6 +820,7 @@ int __isolate_lru_page(struct page *page, int mode, int file)
                 */
                ClearPageLRU(page);
                ret = 0;
+               mem_cgroup_del_lru(page);
        }
 
        return ret;
@@ -1134,7 +1134,6 @@ static unsigned long shrink_inactive_list(unsigned long max_scan,
                        SetPageLRU(page);
                        lru = page_lru(page);
                        add_page_to_lru_list(zone, page, lru);
-                       mem_cgroup_move_lists(page, lru);
                        if (PageActive(page) && scan_global_lru(sc)) {
                                int file = !!page_is_file_cache(page);
                                zone->recent_rotated[file]++;
@@ -1263,7 +1262,7 @@ static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
                ClearPageActive(page);
 
                list_move(&page->lru, &zone->lru[lru].list);
-               mem_cgroup_move_lists(page, lru);
+               mem_cgroup_add_lru_list(page, lru);
                pgmoved++;
                if (!pagevec_add(&pvec, page)) {
                        __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
@@ -2408,6 +2407,7 @@ retry:
 
                __dec_zone_state(zone, NR_UNEVICTABLE);
                list_move(&page->lru, &zone->lru[l].list);
+               mem_cgroup_move_lists(page, LRU_UNEVICTABLE, l);
                __inc_zone_state(zone, NR_INACTIVE_ANON + l);
                __count_vm_event(UNEVICTABLE_PGRESCUED);
        } else {
@@ -2416,6 +2416,7 @@ retry:
                 */
                SetPageUnevictable(page);
                list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list);
+               mem_cgroup_rotate_lru_list(page, LRU_UNEVICTABLE);
                if (page_evictable(page, NULL))
                        goto retry;
        }